|
Download README.md from dcummingsTT/waveformqa: direct link, hf CLI and curl.
- Browser
- Download file 3.77 kB
-
https://huggingface.co/datasets/dcummingsTT/waveformqa/resolve/main/README.md
- Command line
-
hf download hf://datasets/dcummingsTT/waveformqa/README.md
-
curl -L -o README.md https://huggingface.co/datasets/dcummingsTT/waveformqa/resolve/main/README.md
3.77 kB
| license: cc-by-4.0 | |
| pretty_name: WaveformQA VCD Traces | |
| language: | |
| - en | |
| tags: | |
| - vcd | |
| - waveform | |
| - risc-v | |
| - hardware | |
| - rtl-simulation | |
| - temporal-reasoning | |
| - benchmark | |
| size_categories: | |
| - n<1K | |
| viewer: false | |
| # WaveformQA VCD Traces | |
| The 13 VCD traces used by [WaveformQA](https://github.com/tenstorrent/waveformqa), a benchmark for evaluating LLM temporal reasoning over digital signal traces (IEEE ICLAD 2026). The traces were produced by RTL simulation of five open-source RISC-V cores running firmware workloads (~3.3 GB total). | |
| The questions, ground-truth answers, evaluation harness, and per-core generation instructions live in the GitHub repository; this dataset only holds the traces. | |
| ## Files | |
| A single archive, `waveformqa_vcds.zip` (~364 MB; ~3.3 GB extracted), containing a `LICENSE` file (the CC-BY-4.0 terms and third-party attributions below) and one IEEE 1364 VCD file per trace: | |
| | Core | Upstream license | Traces | Signal range | | |
| |------|------------------|--------|--------------| | |
| | [DarkRISCV](https://github.com/darklife/darkriscv) | BSD-3-Clause | `darkriscv_coremark.vcd`, `darkriscv_dhrystone.vcd`, `darkriscv_mandelbrot.vcd` | ~394 | | |
| | [PicoRV32](https://github.com/YosysHQ/picorv32) | ISC | `picorv32_custom.vcd`, `picorv32_dhrystone.vcd` | ~338 | | |
| | [biRISCV](https://github.com/ultraembedded/biriscv) | Apache-2.0 | `biriscv_custom.vcd`, `biriscv_tcm_soc.vcd`, `biriscv_tcm_soc_custom.vcd`, `biriscv_testbench.vcd` | 1,642-2,295 | | |
| | [Ibex](https://github.com/lowRISC/ibex) | Apache-2.0 | `ibex_custom.vcd`, `ibex_dhrystone.vcd` | ~4,185 | | |
| | [SERV](https://github.com/olofk/serv) | ISC | `serv_custom.vcd`, `serv_hello_world.vcd` | ~957 | | |
| ## Usage | |
| Download the archive, extract it into `vcds/` in a clone of the repository, and point the benchmark runner at it: | |
| ```bash | |
| hf download dcummingsTT/waveformqa waveformqa_vcds.zip --repo-type dataset --local-dir vcds | |
| unzip -o vcds/waveformqa_vcds.zip -d vcds | |
| rm vcds/waveformqa_vcds.zip # optional: delete the archive once extracted | |
| uv run python scripts/run_benchmark.py benchmark/t5000 --vcd-dir vcds --eval-only \ | |
| -o output/t5000/evaluation_results.json | |
| ``` | |
| See the repository README (**Getting Started**) for the full set of runner flags. | |
| ## License | |
| Copyright 2026 Tenstorrent USA, Inc. | |
| The VCD traces are generated by simulating the third-party RISC-V cores listed above and are licensed under [CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/legalcode). Signal names and design hierarchy in the traces derive from those cores, which remain under their own licenses. | |
| The ibex_custom and ibex_dhrystone traces derive from the Ibex core. Its upstream NOTICE file reads: | |
| ```text | |
| The Ibex Project | |
| Copyright 2024 lowRISC contributors. | |
| This product includes hardware and/or software developed as part of the | |
| Ibex(R) (https://github.com/lowRISC/ibex) and OpenTitan(R) projects. | |
| Ibex was originally developed by the PULP team at ETH Zurich and University of | |
| Bologna under the name zero-riscy. Ibex verification, performance enhancement | |
| and security hardening have been supported by the OpenTitan project | |
| (https://www.opentitan.org). | |
| ``` | |
| The serv_hello_world trace was produced by running the Zephyr RTOS hello-world image shipped with SERV (Zephyr is Apache-2.0). See [NOTICE](https://github.com/tenstorrent/waveformqa/blob/main/NOTICE) in the repository for full third-party attributions. | |
| ## Citation | |
| ```bibtex | |
| @misc{liu2026waveformqabenchmarkingllmtemporal, | |
| title={WaveformQA: Benchmarking LLM Temporal Reasoning on Digital Waveforms}, | |
| author={Yichuan Liu and Daniel Cummings and Nick Vadlamudi}, | |
| year={2026}, | |
| eprint={2607.20638}, | |
| archivePrefix={arXiv}, | |
| primaryClass={cs.AI}, | |
| url={https://arxiv.org/abs/2607.20638}, | |
| } | |
| ``` | |